Typos of Claude, ChatGPT and other AI tool names lead to malware, Island finds

Mistyping the name of an AI tool can land users on pages that try to get them to install malware, according to research from Island, which makes an enterprise browser. Island traced 186 lookalike domains across 27 AI brands and calls the pattern TypoPrompting.

ChatGPT has the most lookalikes in the set, 25, followed by Claude with 18 and DeepSeek with 16; others imitate Gemini, Grok, Mistral, Perplexity, Midjourney, ElevenLabs, Lovable and Cursor, and some imitate the companies themselves, such as a misspelled OpenAI or Anthropic. One Claude typo led to a ClickFix lure, a fake verification page that gets the visitor to run a hidden PowerShell command. Other typos led to fake "Security Check" pages serving Net Guard, a Microsoft Store-signed app that hijacks the browser and can run remotely supplied scripts.

A third route led to search-hijacking browser extensions built from a shared kit. Island counts 30 such extensions, of which 13 have been removed and 17 are still listed with 972,000 declared users between them. Some of the domains, such as cloude[.]ai, were registered as early as July 2023, The Hacker News reported.